Four layers, four boundaries. A failure in one does not reach the next.

Skinshells Voice in layers Request access
Layer one

Connectivity

The outermost layer, and the one whose failure is total for a site and invisible everywhere else. It answers a single question: can a call happen at all.

In short
  • Keeps the phone system you already own
  • Concurrency from measurement, not headcount
  • Multiple routes, selected per call
Function

What this layer does

It carries signalling and media between your phone system and the public network over IP, in place of a physical circuit. Two things travel: the instructions that set a call up, and the audio itself.

Because the connection is logical, capacity is a configuration value rather than a count of installed channels. That single fact changes the economics — you stop buying blocks and paying for the unused portion indefinitely — and it changes the timescale, because thirty channels become sixty in minutes rather than weeks.

It is normally the first layer fitted, because nothing on anybody's desk changes. The phone on the desk behaves identically; what alters is what sits behind it.

Sub-boundaries

The layers within

Four sub-boundaries, each with its own protection.

Authorisation

Establishing that your phone system is what it claims to be

Address-based for fixed sites, which removes the credential rather than protecting it. Nothing to leak from a configuration backup or a departing administrator.

Transport

Carrying signalling and audio without exposure

TLS for signalling and SRTP for media wherever the endpoint supports them. Where one cannot, the exception is recorded rather than silently accepted.

Selection

Choosing the way out, per call

Multiple upstream paths evaluated on completion rate and audio quality rather than reachability alone. A route that answers and then delivers poor audio is withdrawn.

Containment

Bounding what a compromise can cost

Destinations denied by default, concurrency ceilings enforced, and spend thresholds that stop traffic rather than send an email.

  • Automated scanning finding an exposed registration endpoint within hours
  • Toll fraud initiated outside working hours, exploiting detection latency
  • An upstream route that degrades rather than fails, delivering audio nobody can use
  • A site outage rendering published numbers unreachable with no agreed fallback
  • Capacity exhaustion at peak, presenting as blocked calls rather than as poor audio
Exposure

What this layer protects against

The threats at this boundary are specific and well understood, which is why the defaults are restrictive rather than convenient.

Fitting

How it is fitted

A few weeks, most of it waiting for number ports rather than technical work.

  1. Measure

    Concurrency reconstructed from call records in short intervals across a representative period. Headcount is not an input.

  2. Specify

    Routing, failover destinations and containment thresholds written down and agreed before provisioning.

  3. Prove

    A trial trunk against your phone system exercised with representative call flows, including deliberate failover.

  4. Fit

    Numbers ported on a scheduled date; traffic moved in waves with the previous path retained until verified.

  5. Hold

    After one billing period, measured concurrency compared against the model and the sizing corrected.

Evidence

What this layer produces

Per-call quality records

Jitter, packet loss, round-trip time and completion cause, retained, so a quality dispute becomes a query.

Route health history

Which paths were withdrawn, when and why, available at service review.

Configuration records

Authorised addresses, transport settings and any recorded encryption exception.

Threshold and alert log

What the containment thresholds are set to and every occasion they acted.

Questions

Asked about this layer

All questions answered

No. This layer sits beneath whatever you already have, provided it speaks SIP or can sit behind a gateway. Keeping the existing system is the usual reason this layer is fitted first.

From your call records, reconstructed in short intervals across a representative period, producing a distribution rather than a single figure. Hourly averages suppress the peaks that actually block callers.

Inbound follows the failover agreed at specification — another site, a mobile group or an announcement — invoked automatically. Outbound at that site is unavailable until the circuit returns. The failure does not reach any other site.

Because a route that answers and then delivers unusable audio is worse than one that refuses cleanly. Reachability is a low bar and we do not use it as the health test.

Yes, as a recorded decision attributable to a named person and revisited at the next review. Not as something that quietly happens because a threshold was inconvenient once.

Applications

Where this layer is set differently

Small operations Where the containment boundary matters most, because nobody is watching. Open
Large organisations Where the change boundary carries the most risk. Open
Distributed teams Where the boundary sits outside the organisation entirely. Open
Contact operations Where capacity is a plateau rather than a peak. Open
Next

Ask about this layer specifically.

Name the symptom and the estate. The answer will be about your case rather than about the general one.